How to show up in AI search results in 2026
What improves your chances of AI citations right now, separated from what's untested or already debunked.
By Kuraib AliPublished Updated 13 min read
There's no setting that guarantees a page gets cited by ChatGPT, pulled into an AI Overview, or referenced by Perplexity. Nobody, including Google, claims otherwise. What you can control is whether a page is even eligible to be considered: whether AI systems can reach it, understand it, and have a reason to trust it. That is what I cover here. Everything past that point is a probability, not a switch.
What does "showing up in AI search" mean?
People use "showing up in AI search" to mean five different things, and they're not the same outcome:
- Retrieval. An AI system's search layer finds your page as relevant to a query. This can happen without you ever knowing.
- Consideration. The system pulls your page into the set of sources it's weighing for an answer.
- Citation. The system explicitly names or links your page as a source.
- Mention. Your brand or product is referenced in an answer, with or without a link.
- Inclusion in the generated answer. Information from your page shapes what the AI says, cited or not.
A page can be retrieved constantly and never cited. It can be mentioned by name without ever being linked. Conflating these is why so much "AI search optimisation" advice sounds confident but doesn't hold up. It's usually describing one of these outcomes while implying all five.
Can you optimise for AI search?
Partly, and Google has been direct about the boundary. In its 2026 guidance on generative AI search features, Google states: "From Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO." Its AI Overviews and AI Mode run on the same core Search ranking and quality systems as regular results, built on retrieval-augmented generation: the system retrieves indexed pages and grounds its answer in them rather than generating from memory alone.
That means the foundations that have always mattered still matter: crawlability, indexability, clear and useful content, technical accessibility, and a track record that gives a system reason to trust the source. None of that guarantees a citation. It's the floor a page has to clear before citation is even possible.
1. Make sure AI systems can access your content
Access comes before everything else, and it's the one thing that's purely mechanical to check.
Crawling and robots.txt matter here in a way that's easy to get backwards. AI crawlers split into three different jobs: training, search-indexing, and user-triggered fetching, and each behaves differently. Blocking a training crawler (GPTBot, ClaudeBot) does nothing to your citation chances. Blocking a search-indexing crawler (OAI-SearchBot, Claude-SearchBot, PerplexityBot) is what removes you from consideration. The full breakdown of who's who and the mistakes that trip most sites up is covered in do AI crawlers read your site. Read it before touching robots.txt.
JavaScript rendering is the other access issue, and it's easy to overstate. Vercel's December 2024 analysis of AI-crawler traffic across its network found that none of the major AI crawlers it tested, including OpenAI's and Anthropic's, rendered JavaScript. They read the HTML as delivered, not what a script builds afterwards. That's evidence about those specific crawlers at that point in testing, not a universal law that no AI system can ever render JavaScript. Google's own crawler does render JavaScript, for instance. The safe, current takeaway: if your key facts only appear after a script runs, treat that as a real risk for most non-Google AI systems, and server-side rendering or prerendering is the fix, not a meta tag.
2. Make your content easy to understand
This is a writing and structure problem, not a technical one. A page should make its own purpose obvious in the first few lines: what is this page about, and what specific question does it answer. Descriptive headings that state the actual topic, not clever ones, help both a skimming human and a system trying to extract meaning. Say what things are in plain terms on first use rather than assuming familiarity; a page that spells out "answer engine optimisation (AEO)" once before using the acronym is easier for a system to place correctly than one that doesn't.
None of this is a formatting trick that guarantees anything. It's closer to removing friction. A system that has to guess what a page is about is a system that's less likely to use it confidently.
3. Make sure the right URL can be indexed
Briefly: a page that can't be indexed can't be retrieved, and a page indexed under the wrong URL splits its own signals. If you're running into duplicate URLs, unclear canonical signals, or pages that seem to exist twice, crawling vs indexing covers the mechanics in full.
A canonical tag is a strong hint, not an instruction. Google has documented that it weighs canonical signals (redirects, the rel=canonical tag, and sitemap inclusion) in that order of strength, and can override any of them if the signals conflict or don't match what it observes on the page. Adding a canonical tag doesn't force an outcome; it improves the odds Google agrees with your preference. For the full breakdown of how these signals rank and what to do with each one, see canonical tags and canonicalisation, explained.
4. Publish information worth citing
This is the section that differentiates a page, and it's the one place where there's real research behind the advice rather than repeated assumption.
A 2024 peer-reviewed study by Aggarwal et al. (GEO: Generative Engine Optimization, presented at ACM KDD) tested what changed how often content got used in AI-generated answers. The best methods, citing sources, adding quotations from named sources and adding statistics, raised visibility by up to about 41% on one of the paper's measures and about 28% on another. Keyword stuffing, by contrast, performed roughly 10% worse than doing nothing. The same research found these tactics helped lower-ranked sources the most: a source in fifth place gained the most, while sources already ranked first sometimes lost visibility.
That's a single study, not a settled formula, and it measured what these systems did in that testing environment at that time. Treat it as a strong signal, not a guaranteed lever. What it points towards is intuitive once you see it: specific answers, original expertise, primary sources, and honesty about what you don't know are what separate a page worth citing from one that's technically present but interchangeable with a dozen others. Google's own current guidance says something close to the same thing in different words. It explicitly recommends "creating content that people find unique, compelling, and useful" over almost any technical tactic in this list.
5. Make important answers easy to extract
Once a page has something worth saying, make it easy to lift out. Lead sections with a direct answer before the supporting explanation. Use question-style headings when they match what someone would ask. Use tables and lists where they replace paragraphs of comparison, not as decoration.
What this section deliberately does not recommend is "chunking": breaking content into small artificial fragments on the assumption that AI systems process it that way. Google's current guidance explicitly tells site owners not to rely on this as a special AI-search optimisation requirement, filing it under what it calls "AEO/GEO hacks" alongside creating unnecessary AI-specific text files and pursuing inauthentic mentions. In practice, that means: writing clearly for a human reader already produces content a system can parse just fine. Writing for the system on top of that tends to make the page worse for both.
6. What about structured data?
Schema markup, the structured data that describes a page's content in a machine-readable format, helps search engines understand what a page contains, and it remains useful for earning rich results in regular search. What it does not do, per Google's own current documentation, is guarantee inclusion in AI-generated answers. There's no special "AI schema" that changes your odds. If you already have solid structured data for normal SEO reasons, you're not missing anything AI-specific by not adding more.
7. Do Core Web Vitals matter for AI search?
Google's Core Web Vitals thresholds for 2026 are unchanged: Largest Contentful Paint under 2.5 seconds is "good," Interaction to Next Paint under 200 milliseconds, Cumulative Layout Shift under 0.1, each measured from real-user field data, not a lab score. Google confirms these are used by its ranking systems, but explicitly describes them as one signal among many, not a strong lever on their own; its own guidance says chasing a perfect score "just for SEO reasons may not be the best use of your time."
What Core Web Vitals specifically contribute to AI visibility is less documented. There's no published evidence tying a Core Web Vitals score directly to whether a page gets cited in an AI-generated answer. The reasonable inference, not a confirmed fact, is that Core Web Vitals matter the same way they always have: as part of page experience once a page is indexed. They don't decide whether a page gets indexed. Treat this as background technical health, not an AI-search lever in its own right.
8. SEO vs AEO vs GEO: what's different?
Three terms get used in this space, often as if they're separate disciplines requiring separate strategies:
- SEO: search engine optimisation. Improving a site's visibility in search generally.
- AEO: answer engine optimisation. Structuring content so a system can extract a direct, standalone answer.
- GEO: generative engine optimisation. Making sure a generative AI system can access, trust, and cite a page as a source.
Industry usage of these terms varies, and they're not officially standardised categories with fixed rules. Google's own position, stated plainly in its current guidance, is that from its perspective, optimising for generative AI search "is optimizing for the search experience, and thus still SEO." That doesn't dismiss the underlying shift: AI-mediated discovery is changing how people find information. But it is a clear statement that the fundamentals haven't split into three separate rulebooks. What's different is narrower than the terminology suggests: mainly how content gets accessed (rendering, crawler behaviour) and what makes it citable (evidence, clarity, trust), both covered above.
9. What doesn't reliably make you appear in AI answers
A lot of confidently stated advice in this space isn't backed by anything:
- Keyword stuffing. The Princeton research found this performed worse than doing nothing at all.
- Repeating "AI-optimised" or similar phrases across a site. No documented mechanism connects this to visibility.
- Fake or padded entity signals: naming people, places, or organisations that aren't central to the content, hoping it reads as more "structured."
- Adding FAQ schema purely for AI, with no real need for an FAQ. Schema describes content; it doesn't manufacture relevance that isn't there, and Google's FAQ rich result itself stopped appearing in Search in May 2026.
- Fabricated citations or sources. Beyond the obvious ethical problem, there's no evidence this improves anything, and it directly undermines the trust signals that do matter.
- Publishing large volumes of low-value AI-generated content. Nothing in the evidence gathered here supports volume as a lever; the Princeton findings and Google's own guidance both point towards depth and evidence over quantity.
- Treating llms.txt as a ranking mechanism. Google's current guidance explicitly tells site owners not to rely on this as a special AI-search optimisation requirement. It's filed under the same "AEO/GEO hacks" Google says to skip. That doesn't mean an llms.txt file is harmful to have; it means Google has said specifically that its own systems don't require or reward one. Whether other AI systems make any use of it is a separate, less documented question.
- Blocking or allowing AI crawlers without understanding what each one does. Covered in full in do AI crawlers read your site. The short version is that blanket blocking usually catches the wrong bot.
A practical AI search checklist
| Stage | What to check |
|---|---|
| Crawl | Can search and AI crawlers reach the page? Is robots.txt blocking search/retrieval bots by accident? |
| Index | Is the page indexed under one clear URL, with consistent canonical signals? |
| Understand | Does the page state its purpose and answer clearly, without assuming context the reader doesn't have? |
| Cite | Does it contain something specific (a statistic, a named source, a real example) worth referencing? |
| Trust | Are claims attributable? Is authorship clear? Would a sceptical reader believe this? |
| Measure | What can you observe changing, and over what timeframe? |
None of this guarantees a citation. It gives a page a real chance to be one of the sources considered, which, on the evidence above, is the limit of what you can control.
How do you measure AI search visibility?
This is the weakest part of the current toolkit.
Google's Search Console offers a dedicated Search Generative AI performance report, Google's first official way to see how a site appears inside AI Overviews, AI Mode, and generative Discover features, separate from regular search performance. Google announced it on 3 June 2026 as a limited rollout, and a note on that post says that as of 31 August 2026 Google had "rolled out these insights to all websites worldwide." It has limits:
- It currently reports impressions, how often a page appeared inside these features, broken out by page, country, device, and date.
- It does not currently provide clicks.
- It does not currently provide click-through rate.
- It does not currently provide query-level data.
Alongside the performance report, Google introduced a separate Search generative AI control in Search Console on 3 June 2026, letting a site owner exclude their content from AI Overviews, AI Mode and generative AI features in Discover. Google says the control "isn't used as a ranking or inclusion signal affecting other parts of Search," so using it shouldn't cost a site traditional rankings.
Now that access is global, an empty report in your account is more likely a data-population or property-setup issue than a rollout-eligibility one; give it a few days after checking the right property before assuming something's broken. This section reflects Google's own Search Central blog post and Search Console Help. It is the part of this page I expect to change most as Google develops the report.
Third-party AI-visibility tracking tools exist and can estimate citation frequency across ChatGPT, Perplexity, and other systems by running sample queries and logging results. Treat their numbers as directional, not authoritative. They're sampling a system's live output, not reading Google's or OpenAI's internal data, and results can shift between runs. There is no single, universal "AI visibility score" that any tool, first-party or third-party, currently provides with the reliability of a classic search-ranking report.
Want the crawler mechanics behind §1? See do AI crawlers read your site. For the indexing side of §3, see crawling vs indexing.
Frequently asked questions
What does "showing up in AI search" mean?
People use it to mean five different outcomes that aren't the same: retrieval (your page is found as relevant), consideration (it's pulled into the source set), citation (it's explicitly named or linked), mention (your brand is referenced with or without a link), and inclusion (information from your page shapes the answer, cited or not).
Can you optimise for AI search?
Partly. Google's own 2026 guidance states that optimising for generative AI search is optimising for the search experience, and thus still SEO. The foundations that have always mattered (crawlability, indexability, clear content, and trust) still matter. None of it guarantees a citation; it's the floor a page has to clear before citation is even possible.